People who are not specialists in a particular field often find it difficult to access scientific research. Large language models (LLMs) are capable of summarizing research articles, but are currently unable to automatically modify the tone, depth, and style for different audiences. This thesis introduces SciTeller, a generative artificial intelligence based system designed to transform scientific articles into understandable and engaging stories tailored to specific audiences. Our approach integrates narrative generation and natural language processing. We also propose a taxonomy of eight representative personas, along with a curated dataset of scientific articles and publicly shared stories. Building on this foundation, our system adopts a two-stage pipeline: a Splitter module that extracts and organizes the key content of an article into a structured outline, and a Storyteller module that rewrites each section into a narrative tailored to the selected persona.